About

Lily Hu is an Assistant Professor of Philosophy at Yale University. Her current research projects broadly concern causal theorizing about the social world, with a particular focus on causal inference methodologies in the social sciences. She investigates how various statistical frameworks treat and measure the "causal effect" of social categories such as race, and how these methods are seen to support normative claims about racial discrimination and inequalities more broadly. Previously, she worked on topics in machine learning theory and algorithmic fairness. In addition to her academic work, Lily Hu is a contributing editor at Boston Review and has written for other venues including the Los Angeles Review of Books, Phenomenal World, and The Law & Political Economy Project. She received her doctorate from Harvard University in 2022 and also completed her undergraduate degree in Mathematics at Harvard.

Research topics

• Computer Science
• Political Science
• Artificial Intelligence
• Law
• Sociology
• Machine Learning
• Mathematics
• Environmental ethics
• Mathematical economics
• Economics
• Mathematical optimization
• Psychology
• Philosophy
• Statistics
• Social psychology

Selected publications

• Disparate Treatment and Discriminatory Harm

  SSRN Electronic Journal · 2026-01-01

  preprintOpen accessSenior author
  PublisherDOI

• Making models up

  Philosophy and Phenomenological Research · 2025-09-04

  article1st authorCorresponding

  Abstract Theories of causation within the causal modeling framework are among the most promising and well‐developed approaches to analyzing actual causation today. But since the advent of model‐based theories of causation, authors have struggled with the fact that virtually all such theories issue different causal conclusions when applied to different causal models (of the same case). In recent years, these concerns about model relativity have sharpened into a search for a general theory of model aptness , which lays out principled grounds on which models may be ruled out as inapt for causal analysis. This article presents a new way of understanding what drives model relativity that shows why it is such an enduring problem. I show why extant proposals fail and furthermore, why they are not of the right form to be a solution. Along the way, I draw broader lessons about what an adequate answer to these problems must be like. I then take a step toward such an account by outlining a substantive constraint on when variables and paths in a causal model must not be drawn and when they may.

  PublisherDOI

• Special Issue on New Perspectives on Empirical Methods and Critical Race Theory What Is Perceived When Race Is Perceived and Why It Matters for Causal Inference and Discrimination Studies

  SSRN Electronic Journal · 2025-01-01 · 1 citations

  articleOpen accessSenior author
  PublisherDOI

• Does calibration mean what they say it means; or, the reference class problem rises again

  Philosophical Studies · 2025-05-05 · 1 citations

  article1st authorCorresponding
  PublisherDOI

• Disparate Treatment and Discriminatory Harm

  Legal Theory · 2025-12-01

  articleOpen accessSenior author

  Abstract When do laws and policies that do not explicitly treat people differently on the basis of legally protected traits like race and sex nonetheless constitute disparate treatment on these bases? According to U.S. constitutional law, they do so when “facially neutral” laws are both enacted for impermissible reasons and also produce a discriminatory effect. To date, the first element of this claim – impermissible intention – has attracted significant attention. However, its second element – discriminatory effect – has been largely ignored. Yet it is critical that we better understand what discriminatory effect requires, as competing tests animate debates in Circuit court cases and the issue has recently been flagged by Justice Alito. This Article takes up the task. It explores the normative disagreement that underlies the controversy regarding how to assess whether discriminatory effect is present and diagnoses the genuine moral conflict that any test for discriminatory harm must navigate.

  PublisherOA PDFDOI

• What is New, and What is Old, in Fairness and Machine Learning

  ACM Journal on Responsible Computing · 2025-08-27

  articleOpen access1st authorCorresponding

  This article explores the issue of normative distinctiveness in machine learning decision systems alongside Solon Barocas, Moritz Hardt, and Arvind Narayanan's landmark book Fairness and Machine Learning . What, if anything, is different this time, with the rise of machine learning-based aids to bureaucratic decision-making? I show how a focus on normative distinctiveness can obscure from view a much more significant upshot of machine learning: that the mere existence of feasible alternatives presses new justificatory demands not just on the design of new technical systems but on the prevailing human-centered decision regime. I argue that depoliticizing conventional bureaucratic structures of decision-making leads to a missed opportunity for a broader normative reevaluation of what we owe to each other in a world of expanded practical possibility.

  PublisherOA PDFDOI

• What Is Perceived When Race Is Perceived and Why It Matters for Causal Inference and Discrimination Studies

  Law & Society Review · 2024-12-05 · 3 citations

  articleOpen access1st authorCorresponding

  Abstract Quantifying the causal effects of race is one of the more controversial and consequential endeavors to have emerged from the causal revolution in the social sciences. The predominant view within the causal inference literature defines the effect of race as the effect of race perception and commonly equates this effect with “disparate treatment” racial discrimination. If these concepts are indeed equivalent, the stakes of these studies are incredibly high as they stand to establish or discredit claims of discrimination in courts, policymaking circles and public opinion. This paper interrogates the assumptions upon which this enterprise has been built. We ask: what is a perception of race, a perception of, exactly? Drawing on a rich tradition of work in critical race theory and social psychology on racial cognition, we argue that perception of race and perception of other decision-relevant features of an action situation are often co-constituted; hence, efforts to distinguish and separate these effects from each other are theoretically misguided. We conclude that empirical studies of discrimination must turn to defining what constitutes just treatment in light of the social differences that define race.

  PublisherOA PDFDOI

• What is “Race” in Algorithmic Discrimination on the Basis of Race?

  Journal of Moral Philosophy · 2023-09-05 · 24 citations

  article1st authorCorresponding

  Abstract Machine learning algorithms bring out an under-appreciated puzzle of discrimination, namely figuring out when a decision made on the basis of a factor correlated with race is a decision made on the basis of race . I argue that prevailing approaches, which are based on identifying and then distinguishing among causal effects of race, in their metaphysical timidity, fail to get off the ground. I suggest, instead, that adopting a constructivist theory of race answers this puzzle in a principled manner. On what I call a “thick constructivist” account of race, to be raced is to be socially positioned in the way indicated by a certain set of statistical regularities on the basis of particular phenotypical traits. A thick constructivist sees that acting on the basis of correlations that constitute race qua social position just is acting on the basis of race, because races just are social positions that subject their member individuals to a particular matrix of social relations that define the raced position. This conclusion has considerable ramifications for our understanding of discrimination, algorithms and beyond.

  PublisherDOI

• Rule social media with users: Unveiling Strategical Governance and User Dynamics in Weibo's Community-driven Content Moderation System

  2023-01-09 · 1 citations

  preprintOpen accessSenior author

  Social media companies continuously experiment with varied platform governance models to tackle content moderation challenges, which calls for a comprehensive and empirical understanding of how a content moderation system evolves on a long-term scale. Our study aims to fill this gap with a quantitative and qualitative study of Weibo's community-driven content moderation system with policy documents and eleven million public moderation cases and decision data from 2012 to 2021. Based on the reporting activities, platform decisions, and jury actions, we investigated the motivations and behavior patterns of three important actors in this governance model: reporting users, platform authority, and user jurors. We suggest that users who frequently reported content and initiated the community-driven content moderation process, usually have a pattern of voluntarily policing the community or abusing others, sometimes coordinately, and were also treated differently by the platform. We indicate that Weibo's strategical moderation decisions have significantly distinctive preferences over cases from different categories or with different harmfulness levels, and the cases involving socially sensitive issues were given more consideration and penalized more severely than common misbehavior. We explore how Weibo authority leveraged the usually one-sided votes of digital jurors to endorse its final decisions while the reason notes given by crowdsourced jurors also revealed a serious issue of decaying motivation. Our findings offer important insights into understanding the coordination between a social media platform and voluntary users to moderate an online community, and resonate with the question of how an autonomic platform governance model can prevail or perish.

  PublisherOA PDFDOI

• Fault Detection and Diagnostics for HVAC&R Equipment*

  River Publishers eBooks · 2021-01-07

  book-chapter

  This chapter provides an overview of fault detection and diagnostics (FDD), including descriptions of fundamental processes, important definitions, and examples that building operators and managers can implement using data collected from the building automation systems or dedicated logging devices. Poorly maintained, degraded, improperly controlled equipment wastes an estimated 10% to 30% of the energy used in commercial buildings. Much of this waste could be prevented with widespread adoption of FDD, an area of investigation concerned with automating the processes of detecting faults in physical systems and diagnosing their causes. Fault detection and diagnostics can be performed “manually” through visual inspection of charts, trends or can be fully automated. In addition to the data, the basic building blocks of automated FDD systems are the methods for detecting faults and diagnosing their causes. Approaches to FDD range from methods based on physical, analytical models based entirely on first principles, to those driven by performance data and using artificial intelligence or statistical techniques.

  PublisherDOI

Frequent coauthors

• Dennis F. Deen

  Neurological Surgery

  59 shared

• Kathleen R. Lamborn

  47 shared

• Jingli Wang

  Medical College of Wisconsin

  21 shared

• Tomoko Ozawa

  University of California, San Francisco

  18 shared

• Hangjun Ruan

  University of California, San Francisco

  16 shared

• Andrew W. Bollen

  University of California, San Francisco

  13 shared

• Krisztina Pongracz

  Menlo School

  9 shared

• Sergei Gryaznov

  9 shared